中文摘要：

针对高光谱影像处理应用中,标记样本往往数量较小且质量不均而未标记样本大量存在的问题,结合半监督学习方法,提出一种面向高光谱影像分类的半监督极限学习机分类算法.首先根据图理论,联合高光谱影像空间光谱信息,对标记和未标记样本共同构建无向加权图;然后,考虑平滑性约束和结构最小化原则,构造分类目标函数;最后,利用核方法求解最优参数,进而实现高光谱影像的半监督分类.采用该方法进行分类对比实验,结果表明：该方法能够有效利用未标记样本信息,提高小样本下的高光谱影像分类精度.

英文摘要：

For hyperspectral imagery＇s processing, the number of labeled samples is often small, and the quality of them is uneven, and there exist a large number of unlabeled samples. Aiming at this problem, a semi-supervised algorithm based on extreme learning machine for hyperspectral imagery classification was presented. Firstly, according to the theory of the graph, the undirected weighted graph was constructed, and the graph was combined with both labeled and unlabeled samples spectral and spatial features. Then, by considering the smoothness constraint and the structure minimization principle, the classification objective function was constructed. Finally, parameters were solved and semi-supervised classification of hyperspectral image was achieved. The experimental results show that the proved method can improve the classification accuracy effectively by using unlabeled samplesp information when the labeled samplesp size is small.